Rethinking Underwriting: Improved Scores, New Scores, and Other New Aspects of Underwriting Will Provide Finer and More Granular Risk Discrimination in a Given Pool of Applicants. Institutions Will Continue to Rely on Automated Decision Processes in Order to Contain Cost per Loan
Carroll, Peter, The RMA Journal
Now, as the economy starts to emerge from recession, is an excellent time for lenders to rethink and upgrade their lending processes.
In late 2008 and early 2009, survival was the primary goal of many lending institutions. But these days, banks that survived the financial crisis are cautiously turning to the future, knowing they must restart their new account acquisition efforts to compete effectively in a smaller, lower-growth market.
Naturally, lenders are determined not to repeat the mistakes of the past few years. Hence, the growing need to "rethink underwriting." And this will mean much more than just loosening or tightening credit standards.
One way to get started is to pick apart what just happened to consumer credit markets, see what lessons can be learned, and decide what to do next.
Debate continues about the root causes of the financial crisis and the recession that followed it. Whatever history decrees to have been the causes, the following were certainly major effects:
* Loan portfolios defaulted at far higher rates than were anticipated.
* Loss severity, especially on real estate loans, ran far higher than expected.
* The ratio of actual to expected default rates varied widely by product and credit-score band (Figure 1), though higher-quality borrowers showed the highest ratios.
In short, despite the seeming sophistication of underwriting methods developed over the past 30 years and the implementation of bank-wide risk and economic capital frameworks under Basel I and II, banks were hit with losses that were almost entirely unanticipated.
When losses soared in 2008 and 2009, many lenders reverted to human underwriting. In a series of recent client interviews, Oliver Wyman found that about 90% of large banks that make small business loans had either abandoned score-based adjudication processes in favor of human underwriters, or had severely cut back the proportion of loans that were auto-adjudicated using a score matrix of the business and the owner.
There was an element of panic about these changes of policy, of course, but lenders were convinced they needed to reestablish a human role in decision making. Relying on human underwriters again, insisting on more extensive application information, and conducting site visits for small business loans all combined to cut back underwriting capacity severely. Far fewer loans per week could be adjudicated and the cost per loan was much higher than before. These issues didn't matter at the time, however, because the banks told us they were seeking and approving very few applicants anyway.
What Lessons Can Be Learned?
Credit Scores Can Be Upgraded
As losses skyrocketed, one lesson that appeared to emerge was that credit scores had proven to be a failure. Lenders had increasingly relied on credit scores in their underwriting, and now the loans they had approved were defaulting in droves. Actual default rates were far higher than the good/bad odds associated with the scores in use when the loans were booked.
Upon closer examination, however, the demise of credit scores had been greatly exaggerated. Credit scores actually performed quite well in terms of rank-ordering risk. Consumer credit scores are, first and foremost, a device for indicating relative likelihood of default; however, scores are also typically calibrated by users so that a given score at loan approval can be associated with a specific default probability.
What happened in the last three years--as shown by Figure 1--is that the relative payment/default performance of borrowers tracked very well with their score bands: People with higher scores indeed defaulted at a lower rate than those with lower scores. But the calibration turned out to be way off: Actual default rates far exceeded expected default rates across all score bands. …